# Comparison of Methods for Estimating Fractional Cover of Photosynthetic and Non-Photosynthetic Vegetation in the Otindag Sandy Land Using GF-1 Wide-Field View Data

^{1}

^{2}

^{3}

^{*}

## Abstract

**:**

_{pv}) and NPV (f

_{npv}) using multispectral information. The purpose of this study is to evaluate several spectral unmixing approaches for retrieval of f

_{pv}and f

_{npv}in the Otindag Sandy Land using GF-1 wide-field view (WFV) data. To deal with endmember variability, pixel-invariant (Spectral Mixture Analysis, SMA) and pixel-variable (Multi-Endmember Spectral Mixture Analysis, MESMA, and Automated Monte Carlo Unmixing Analysis, AutoMCU) endmember selection approaches were applied. Observed fractional cover data from 104 field sites were used for comparison. For f

_{pv}, all methods show statistically significant correlations with observed data, among which AutoMCU had the highest performance (R

^{2}= 0.49, RMSE = 0.17), followed by MESMA (R

^{2}= 0.48, RMSE = 0.21), and SMA (R

^{2}= 0.47, RMSE = 0.27). For f

_{npv}, MESMA had the lowest performance (R

^{2}= 0.11, RMSE = 0.24) because of coupling effects of the NPV and bare soil endmembers, SMA overestimates f

_{npv}(R

^{2}= 0.41, RMSE = 0.20), but is significantly correlated with observed data, and AutoMCU provides the most accurate predictions of f

_{npv}(R

^{2}= 0.49, RMSE = 0.09). Thus, the AutoMCU approach is proven to be more effective than SMA and MESMA, and GF-1 WFV data are capable of distinguishing NPV from bare soil in the Otindag Sandy Land.

## 1. Introduction

## 2. Materials and Methods

#### 2.1. Study Area

#### 2.2. Data Used in this Study

#### 2.2.1. Remote Sensing Data

#### 2.2.2. Field Spectroscopy

^{®}4 spectroradiometer with a 25° sensor foreoptic. All measurements were collected within two hours of local solar noon on clear sky days. The sensor was held 1 m above the top of the PV and NPV canopy or bare soil surface in a vertical downward position. Prior to each measurement, the spectroradiometer was calibrated to a Spectralon

^{®}white reference target.

#### 2.2.3. Fractional Ground Cover Data

#### 2.3. Methods

#### 2.3.1. Spectral Mixture Analysis

#### 2.3.2. Multiple Endmember SMA

#### 2.3.3. AutoMCU

#### 2.3.4. Unmixing Strategy

#### 2.3.5. Comparison with Observed Data

^{2}) of linear regression:

## 3. Results

#### 3.1. Field f_{pv} f_{npv}, and f_{soil} Measurements

_{pv}, f

_{npv}, and f

_{soil}followed the expected temporal and structural patterns (Table 2). Because the acquiring time corresponds to the maximum vegetation growing season, fields were dominated by green vegetation resulting in high f

_{pv}. Even so, NPV takes an important proportion of sample plots, especially in open woodland (average f

_{npv}reaches 24%). Compared to grassland, the open woodland was characterized by higher f

_{npv}and lower f

_{pv}, while no obvious difference was observed for f

_{soil}.

#### 3.2. Endmember Library and Variability

#### 3.3. Fractional Cover Estimation and Validation

#### 3.3.1. SMA

_{pv}, f

_{npv}, and f

_{soil}in the Otindag Sandy Land were estimated using the FCLS algorithm. The RMSE of SMA for the samples ranges from 1.24% to 19.91%, with an average RMSE value of 3.76%. The spatial distribution of f

_{pv}and f

_{npv}is shown in Figure 4a. Overall, the spatial distribution characteristics are consistent with prior field investigations. From east to west, the fractional cover of both PV and NPV shows an obvious decreasing tendency, coincident with decreasing precipitation. The high f

_{pv}, mainly located in the eastern part of the Otindag Sandy Land, corresponds to forest, farmland, and wetland. The high f

_{npv}, was mainly found in the grassland along the southern and northern edge of the eastern Otindag Sandy Land, and the high f

_{soil}mainly correspond to desert steppe and active sand dunes, respectively.

_{pv}and f

_{npv}are shown in Figure 5, and their statistical data are summarized in Table 3. It can be seen that the SMA-based f

_{pv}and f

_{npv}are highly correlated with observed data, with R

^{2}values of 0.47 and 0.41, respectively. However, there is an obvious underestimation of f

_{pv}(RMSE = 0.27), and an obvious overestimation of f

_{npv}(RMSE = 0.20). The high correlation demonstrates that PV and NPV can be distinguished by SMA using GF-1 WFV data, but the absolute error suggests that using the average spectra of the PV, NPV, and bare soil endmember library as the invariant endmember is problematic.

#### 3.3.2. MESMA

_{pv}, f

_{npv}and f

_{soil}is shown in Figure 4b. The overall spatial pattern is consistent with the SMA-based results, but with obvious difference of f

_{pv}, f

_{npv}in the east Otindag Sandy Land. The minimal RMSE was selected as the criterion to choose the appropriate endmember from the endmember library. The RMSEs of MESMA for the sample plots range from 0% to 8.46%, with an average RMSE value of 0.53%, significantly lower than that of the SMA (3.76%), indicating that MESMA fits the GF-1 WFV data better than SMA.

_{pv}correlate with observed f

_{pv}well (with R

^{2}of 0.48), but with the obvious underestimation of f

_{pv}(RMSE = 0.21). However, the accuracy of MESMA-based f

_{npv}and f

_{soil}are poor, with a weak relationship with observed data (with R

^{2}of 0.11 and 0.15, respectively). Therefore, MESMA performs better for f

_{pv}, because the PV endmember is clearly different from the NPV and bare soil endmembers, but would lead to confusion with regard to f

_{npv}and f

_{soil}because of the similarities of the NPV and bare soil endmember spectra.

#### 3.3.3. AutoMCU

_{pv}and f

_{npv}distribution and their average values along unmixing times obey the same change rules, one of which is illustrated in Figure 6. The f

_{pv}and f

_{npv}values are highly variable during the first 50 iterations, indicating that endmember selection evidently influences f

_{pv}and f

_{npv}estimation results. When the number of iterations exceeds 150, the mean f

_{pv}and f

_{npv}values are essentially stable. In addition, the f

_{pv}and f

_{npv}distributions follow Gaussian distribution, indicating that their expected (average) values represent the true values. Therefore, 150 iterations were used for unmixing of the entire GF-1 WFV dataset by AutoMCU.

_{pv}, f

_{npv}, and f

_{soil}show that AutoMCU performs well for f

_{pv}and f

_{npv}. The R

^{2}of the regression of f

_{pv}and f

_{npv}on observed data is 0.49, indicating a significant correlation with observed data. The RMSEs for the f

_{pv}and f

_{npv}estimation are 0.17 and 0.09, respectively.

#### 3.3.4. Comparisons of Different Methods

_{pv}was improved compared to SMA (increase in R

^{2}from 0.47 to 0.48, and decrease in RMSE from 0.27 to 0.21), and the underestimation of f

_{pv}in SMA was alleviated slightly. However, the accuracy of MESMA-based f

_{npv}deteriorates significantly, with a weak relationship with observed data (R

^{2}= 0.11) and an increase in the RMSE from 0.22 to 0.24. In addition, the MESMA-based f

_{npv}was significantly overestimated, similar to SMA. Therefore, MESMA performs better for estimating f

_{pv}, but would lead to serious mistakes for f

_{npv}and f

_{soil}. Therefore, AutoMCU is an effective approach for f

_{pv}and f

_{npv}estimation in the Otindag Sandy Land because it performs better than SMA and MESMA.

_{pv}and f

_{npv}estimation are different for different land cover types. For example, in f

_{pv}estimation with the AutoMCU approach, grassland has the highest RMSE (0.19), followed by grassland encroached by shrub (RMSE = 0.16) and open woodland (RMSE = 0.14). However, the opposite tendency appears in f

_{npv}estimation. Open woodland has the highest RMSE (0.13), followed by grassland encroached by shrub (RMSE = 0.10), and grassland (RMSE = 0.07). Differences in accuracy were consistent with differences in average f

_{pv}and f

_{npv}values of sample plots in different land cover types (Table 3).

## 4. Discussion

#### 4.1. Separability of NPV and Bare Soil in SMA

_{npv}and f

_{soil}using the GF-1 WFV data is the determination of appropriate endmembers—especially the bare soil endmember, which shows large intra-variability when compared to NPV spectral libraries—and not the absolute difference of reflectance of NPV and bare soil. Thus, identifying the most appropriate NPV and bare soil endmember for unmixing is critical for retrieving ground cover accurately.

#### 4.2. Effects of Endmember Variability

_{pv}and f

_{npv}in the Otindag Sandy Land with GF-1 WFV data. Table 2 shows that all SMA-based fractions are highly correlated with observed data, but that f

_{pv}is clearly underestimated, while f

_{npv}is overestimated; only f

_{soil}shows no directional bias. Thus, the use of the average spectra of PV and NPV as invariant endmembers is not appropriate, but identifying the most representative PV and NPV endmember spectra is extremely difficult.

_{pv}and f

_{npv}using Landsat [33], MODIS [34] and, Geoeye [35] data. However, MESMA was shown to be incapable for f

_{npv}estimation with GF-1 WFV data in the Otindag Sandy Land, because it would lead to serious confusion between NPV and bare soil, with R

^{2}values of 0.11 and 0.15, and RMSEs of 0.24 and 0.21, respectively. The endmembers most frequently utilized in MESMA are shown in Figure 7. Three PV (including mean spectra), seven NPV, and four bare soil endmembers were utilized most frequently, accounting for 70%, 68%, and 93%, respectively. From the spectral curve, we can see that the bare soil endmember spectra differ more widely than those for PV and NPV. Thus, the selection of the bare soil endmember will have a significant effect on the unmixing results. For example, the use of the bare soil endmember with obviously higher reflectance would lead to a significant underestimation of f

_{soil}. Therefore, the appropriate bare soil endmember determination would be essential for estimating f

_{npv}and f

_{soil}accurately. Since the spectral similarity between NPV and bare soil, the actual NPV and bare soil endmember combinations could be replaced by another combinations in MESMA in order to minimize the RMSE of spectral fit. This finding is consistent with Okin et al. [26], who found that MESMA did not improve f

_{soil}estimates, because the similarity between soil and NPV spectra can actually lead to errors in MESMA as some bare soil/NPV combinations may be mistaken as combinations of other bare soil/NPV [36]. Therefore, it cannot be assumed that MESMA always leads to better f

_{npv}and f

_{soil}estimates.

_{pv}and f

_{npv}estimation. Like in previous studies, the AutoMCU is shown to be effective in semi-arid sandy lands. In addition to improving computational efficiency, AutoMCU is able to explicitly quantify the proportion indeterminacy [40]. It is worth noting that the AutoMCU was originally designed for hyperspectral sensors [25], and that only few studies applied it to multispectral sensors, mainly Landsat [37,41], but with a lack of quantitative validation of f

_{pv}and f

_{npv}. The distinguishing feature of this study is that we were able to show that GF-1 WFV data, lacking the short-wave infrared bands, are capable of distinguishing NPV from bare soil in the Otindag Sandy Land.

_{pv}and f

_{npv}are shown in Figure 8. Compared to SMA, AutoMCU-based f

_{pv}became higher (up to 10%) in east Otindag Sandy Land, except for cropland and wetland with the highest green vegetation cover, and decreased (up to 16%) in the entire west Otindag Sandy Land, corresponding to desert steppe; AutoMCU-based f

_{npv}decreased for most vegetated areas (up to 20%), resolving the overestimation problem of f

_{npv}in the SMA (Figure 5). However, there are extensive regions with increasing f

_{npv}(up to 10%), which were verified to correspond to regions with a bright background of sandy soil. More importantly, these increases do not lead to more serious overestimation. Thus, f

_{npv}estimation accuracy was improved effectively. Compared to the MESMA, AutoMCU-based f

_{pv}and f

_{npv}changed dramatically, because MESMA led to a serious confusion between NPV and bare soil. f

_{pv}increased for vegetated areas and decreased for sandy lands, which alleviated the underestimation problem of f

_{pv}in MESMA. f

_{npv}decreased significantly for salt marshes in west regions, which was misclassified as NPV in MESMA.

#### 4.3. Cross-Multispectral Sensor Comparison

_{pv}estimations are relatively stable, regardless of the method used. Estimated f

_{pv}is significantly correlated with field observed f

_{pv}(p < 0.0001), with R

^{2}values ranging from 0.47 to 0.49. In addition, the obvious underestimation problem gradually improved from SMA to MESMA and AutoMCU, which shows that the per-pixel variable endmember combinations will increase the accuracy of the f

_{pv}estimation. However, the f

_{npv}estimations are highly variable: the SMA- and AutoMCU-based f

_{npv}estimates are significantly correlated with field observed f

_{npv}, with R

^{2}values of 0.41 and 0.49, respectively, but, the correlation between MESMA-based f

_{npv}and in situ f

_{npv}is poor (R

^{2}= 0.11).

_{pv}and f

_{npv}. Here, f

_{npv}is estimated most accurately, with an RMSE of 0.09 (R

^{2}= 0.49), while f

_{pv}is estimated less accurately, with an RMSE of 0.17 (R

^{2}= 0.49), and f

_{soil}shows the worst estimation results, with an RMSE of 0.20 (R

^{2}= 0.48). An important aspect of this study is the field verification. Few studies compare f

_{pv}and f

_{npv}from multispectral remote sensing imagery with field measurements. Table 4 [26,42,43] lists recently published results for f

_{pv}and f

_{npv}estimation based on different multispectral sensors. Since source data, study region, and study period differ, the comparison to our study is limited. For f

_{pv}estimation, this study shows lower accuracy, with the lowest RMSE acquired by AutoMCU of 17%, while the RMSE numbers of previous studies range from 7% to 14.7%. However, this study demonstrates improved f

_{npv}estimation: the lowest RMSE acquired by AutoMCU is 9%, compared to the previous studies’ 12%–20.5%. An important difference is that the accuracy for f

_{npv}estimation is higher than that for f

_{pv}estimation, which might partly be caused by the lower field observed f

_{npv}coverage (≤45%) during the peak growing season in our study area, while field observed f

_{npv}values ranged from 0% to 100% when multi-temporal data were considered. Our main goal was to validate the ability of the four-band GF-1 WFV (not including any SWIR bands) to distinguish PV and NPV from bare soil. Based on f

_{pv}and f

_{npv}estimation accuracy comparisons, we conclude that both f

_{pv}and f

_{npv}can be determined with modest accuracy using GF-1 WFV data with the AutoMCU approach.

#### 4.4. Uncertainties, and Sources of Error

_{pv}and f

_{npv}estimation errors could be partially related to errors in the field measurements. Sampling procedure along the diagonal lines has been widely acknowledged to be effective for ground cover measurement of sample plots, wherein vegetation is not distributed in parallel rows [44,45]. In particular, field surveyors were well trained and measurements of each transect for the first 20 sample plots were cross-validated to guarantee reliability of the field reference data. However, there were some errors caused by the observer’s bias. It is most difficult to acquire consistent field data for NPV and estimates can vary greatly between observers. This can result in the confusion between PV and NPV.

## 5. Conclusions

_{pv}and f

_{npv}based on GF-1 WFV data without any SWIR bands. The unmixing results were evaluated against observed data collected concurrently with image acquisition. The main findings are:

- (1)
- Despite spectral similarity of NPV and bare soil, there are some differences in the GF-1 WFV bands. First and foremost, the bare soil spectra are significantly higher than the NPV spectra, mainly due to extensive bright sandy substrate. In addition, a bow-shaped protuberance exists from blue to red bands in the bare soil spectra, which is not present in the NPV spectra.
- (2)
- Due to the complex and bright soil background of the Otindag Sandy Land, the bare soil endmember libraries show large intra-variability. Therefore, determining the appropriate endmember combinations, especially the bare soil endmember, is a key process for successfully estimating f
_{pv}and f_{npv}. - (3)
- Invariant endmember combinations should be used with caution, because they can lead to serious over- or underestimation problems (SMA). The MESMA cannot be assumed to always perform better than SMA, due to the coupling of the NPV and bare soil endmembers. AutoMCU was shown to be effective for dealing with endmember variability while acquiring accurate f
_{pv}and f_{npv}estimation. Compared with SMA, both R^{2}and RMSE are improved. - (4)
- Compared to other relevant multispectral applications, the GF-1 WFV data were shown to be capable for f
_{pv}and f_{npv}estimation in the Otindag Sandy Land, despite a lack of the important SWIR bands, which are considered important for separation of NPV from bare soil. With GF-1 WFV’s unique advantage of high spatial resolution (16 m), wide coverage (800 km), and high revisit frequency (2–3 days), there is great potential for future analyses.

## Acknowledgments

## Author Contributions

## Conflicts of Interest

## References

- Ruiz-Colmenero, M.; Bienes, R.; Eldridge, D.; Marques, M. Vegetation cover reduces erosion and enhances soil organic carbon in a vineyard in the central Spain. Catena
**2013**, 104, 153–160. [Google Scholar] [CrossRef] - Higginbottom, T.P.; Symeonakis, E. Assessing land degradation and desertification using vegetation index data: Current frameworks and future directions. Remote Sens.
**2014**, 6, 9552–9575. [Google Scholar] [CrossRef] - Asner, G.P.; Heidebrecht, K.B. Imaging spectroscopy for desertification studies: Comparing aviris and eo-1 hyperion in argentina drylands. IEEE Trans. Geosci. Remote Sens.
**2003**, 41, 1283–1296. [Google Scholar] [CrossRef] - Bastin, G.; Scarth, P.; Chewings, V.; Sparrow, A.; Denham, R.; Schmidt, M.; O’Reagain, P.; Shepherd, R.; Abbott, B. Separating grazing and rainfall effects at regional scale using remote sensing imagery: A dynamic reference-cover method. Remote Sens. Environ.
**2012**, 121, 443–457. [Google Scholar] [CrossRef] - Haijiang, L.; Chenghu, Z.; Weiming, C.; En, L.; Rui, L. Monitoring sandy desertification of otindag sandy land based on multi-date remote sensing images. Acta Ecol. Sin.
**2008**, 28, 627–635. [Google Scholar] [CrossRef] - Wu, B.; Li, X.; Liu, W.; Yang, X.; Lu, Q. Desertification control regionalization and rehabilitation counter meatures of source area of sand and dust endangering Beijing–Tianjin. Scientia Silave Sinicae
**2006**, 42, 65–70. [Google Scholar] - Wu, Z.; Wu, J.; Liu, J.; He, B.; Lei, T.; Wang, Q. Increasing terrestrial vegetation activity of ecological restoration program in the Beijing–Tianjin sand source region of China. Ecol. Eng.
**2013**, 52, 37–50. [Google Scholar] [CrossRef] - Li, X.; Zhang, J. Derivation of the green vegetation fraction of the whole China from 2000 to 2010 from MODIS data. Earth Interact.
**2016**, 20, 1–16. [Google Scholar] [CrossRef] - Jiang, Z.; Huete, A.R.; Chen, J.; Chen, Y.; Li, J.; Yan, G.; Zhang, X. Analysis of NDVI and scaled difference vegetation index retrievals of vegetation fraction. Remote Sens. Environ.
**2006**, 101, 366–378. [Google Scholar] [CrossRef] - Gitelson, A.A.; Kaufman, Y.J.; Stark, R.; Rundquist, D. Novel algorithms for remote estimation of vegetation fraction. Remote Sens. Environ.
**2002**, 80, 76–87. [Google Scholar] [CrossRef] - Tucker, C.J. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sens. Environ.
**1979**, 8, 127–150. [Google Scholar] [CrossRef] - Huete, A.; Didan, K.; Miura, T.; Rodriguez, E.P.; Gao, X.; Ferreira, L.G. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ.
**2002**, 83, 195–213. [Google Scholar] [CrossRef] - Okin, G.S. The contribution of brown vegetation to vegetation dynamics. Ecology
**2010**, 91, 743–755. [Google Scholar] [CrossRef] [PubMed] - Guerschman, J.P.; Hill, M.J.; Renzullo, L.J.; Barrett, D.J.; Marks, A.S.; Botha, E.J. Estimating fractional cover of photosynthetic vegetation, non-photosynthetic vegetation and bare soil in the Australian tropical savanna region upscaling the eo-1 hyperion and MODIS sensors. Remote Sens. Environ.
**2009**, 113, 928–945. [Google Scholar] [CrossRef] - Elmore, A.J.; Asner, G.P.; Hughes, R.F. Satellite monitoring of vegetation phenology and fire fuel conditions in Hawaiian Drylands. Earth Interact.
**2005**, 9, 1–21. [Google Scholar] [CrossRef] - Okin, G.S. Relative spectral mixture analysis—A multitemporal index of total vegetation cover. Remote Sens. Environ.
**2007**, 106, 467–479. [Google Scholar] [CrossRef] - Aguilar, J.; Evans, R.; Daughtry, C. Performance assessment of the cellulose absorption index method for estimating crop residue cover. J. Soil Water Conserv.
**2012**, 67, 202–210. [Google Scholar] [CrossRef] - Daughtry, C.S.; Doraiswamy, P.; Hunt, E., Jr.; Stern, A.; McMurtrey, J., III; Prueger, J. Remote sensing of crop residue cover and soil tillage intensity. Soil Tillage Res.
**2006**, 91, 101–108. [Google Scholar] [CrossRef] - Ren, H.; Zhou, G.; Zhang, F.; Zhang, X. Evaluating cellulose absorption index (CAI) for non-photosynthetic biomass estimation in the desert steppe of Inner Mongolia. Chin. Sci. Bull.
**2012**, 57, 1716–1722. [Google Scholar] [CrossRef] - Jia, G.J.; Burke, I.C.; Goetz, A.F.; Kaufmann, M.R.; Kindel, B.C. Assessing spatial patterns of forest fuel using AVIRIS data. Remote Sens. Environ.
**2006**, 102, 318–327. [Google Scholar] [CrossRef] - Huete, A.R.; Miura, T.; Gao, X. Land cover conversion and degradation analyses through coupled soil-plant biophysical parameters derived from hyperspectral EO-1 hyperion. IEEE Trans. Geosci. Remote Sens.
**2003**, 41, 1268–1276. [Google Scholar] [CrossRef] - Qi, J.; Wallace, O. Biophysical attributes estimation from satellite images in arid regions. In Proceedings of the 2002 IEEE International Geoscience and Remote Sensing Symposium (IGARSS’02), Toronto, ON, Canada, 24–28 June 2002; pp. 2000–2002.
- Marsett, R.C.; Qi, J.; Heilman, P.; Biedenbender, S.H.; Watson, M.C.; Amer, S.; Weltz, M.; Goodrich, D.; Marsett, R. Remote sensing for grassland management in the arid southwest. Rangel. Ecol. Manag.
**2006**, 59, 530–540. [Google Scholar] [CrossRef] - Cao, X.; Chen, J.; Matsushita, B.; Imura, H. Developing a MODIS-based index to discriminate dead fuel from photosynthetic vegetation and soil background in the asian steppe area. Int. J. Remote Sens.
**2010**, 31, 1589–1604. [Google Scholar] [CrossRef] - Asner, G.P.; Heidebrecht, K.B. Spectral unmixing of vegetation, soil and dry carbon cover in arid regions: Comparing multispectral and hyperspectral observations. Int. J. Remote Sens.
**2002**, 23, 3939–3958. [Google Scholar] [CrossRef] - Okin, G.S.; Clarke, K.D.; Lewis, M.M. Comparison of methods for estimation of absolute vegetation and soil fractional cover using MODIS normalized BRDF-adjusted reflectance data. Remote Sens. Environ.
**2013**, 130, 266–279. [Google Scholar] [CrossRef] - Muir, J. Field Measurement of Fractional Ground Cover: A Technical Handbook Supporting Ground Cover Monitoring for Australia; ABARES: Canberra, Australia, 2011.
- Roberts, D.A.; Gardner, M.; Church, R.; Ustin, S.; Scheer, G.; Green, R. Mapping chaparral in the Santa Monica Mountains using multiple endmember spectral mixture models. Remote Sens. Environ.
**1998**, 65, 267–279. [Google Scholar] [CrossRef] - Asner, G.P.; Lobell, D.B. A biogeophysical approach for automated SWIR unmixing of soils and vegetation. Remote Sens. Environ.
**2000**, 74, 99–112. [Google Scholar] [CrossRef] - Heinz, D.C.; Chang, C.-I. Fully constrained least squares linear spectral mixture analysis method for material quantification in hyperspectral imagery. IEEE Trans. Geosci. Remote Sens.
**2001**, 39, 529–545. [Google Scholar] [CrossRef] - Chang, C.-I.; Heinz, D.C. Constrained subpixel target detection for remotely sensed imagery. IEEE Trans. Geosci. Remote Sens.
**2000**, 38, 1144–1159. [Google Scholar] [CrossRef] - Gao, Z.; Bai, L.; Wang, B.; Li, Z.; Li, X.; Wang, Y. Estimation of soil organic matter content in desertified lands using measured soil spectral data. Scientia Silvae Sinicae
**2011**, 47, 9–16. [Google Scholar] - Numata, I.; Roberts, D.A.; Chadwick, O.A.; Schimel, J.; Sampaio, F.R.; Leonidas, F.C.; Soares, J.V. Characterization of pasture biophysical properties and the impact of grazing intensity using remotely sensed data. Remote Sens. Environ.
**2007**, 109, 314–327. [Google Scholar] [CrossRef] - Ballantine, J.-A.C.; Okin, G.S.; Prentiss, D.E.; Roberts, D.A. Mapping north african landforms using continental scale unmixing of MODIS imagery. Remote Sens. Environ.
**2005**, 97, 470–483. [Google Scholar] [CrossRef] - Mishra, N.B.; Crews, K.A. Estimating fractional land cover in semi-arid central kalahari: The impact of mapping method (spectral unmixing vs. Object-based image analysis) and vegetation morphology. Geocarto Int.
**2014**, 29, 860–877. [Google Scholar] [CrossRef] - Okin, G.S.; Roberts, D.A.; Murray, B.; Okin, W.J. Practical limits on hyperspectral vegetation discrimination in arid and semiarid environments. Remote Sens. Environ.
**2001**, 77, 212–225. [Google Scholar] [CrossRef] - Davidson, E.A.; Asner, G.P.; Stone, T.A.; Neill, C.; Figueiredo, R.O. Objective indicators of pasture degradation from spectral mixture analysis of Landsat imagery. J. Geophys. Res. Biogeosci.
**2008**, 113. [Google Scholar] [CrossRef] - Asner, G.P.; Levick, S.R.; Smit, I.P. Remote sensing of fractional cover and biochemistry in savannas. In Ecosystem Function in Savannas: Measurement and Modelling at Landscape to Global Scales; CRC Press: Boca Raton, FL, USA, 2011; pp. 195–218. [Google Scholar]
- Harris, A.T.; Asner, G.P. Grazing gradient detection with airborne imaging spectroscopy on a semi-arid rangeland. J. Arid Environ.
**2003**, 55, 391–404. [Google Scholar] [CrossRef] - Zare, A.; Ho, K. Endmember variability in hyperspectral analysis: Addressing spectral variability during spectral unmixing. IEEE Signal Proc. Mag.
**2014**, 31, 95–104. [Google Scholar] [CrossRef] - Asner, G.P.; Knapp, D.E.; Cooper, A.N.; Bustamante, M.M.; Olander, L.P. Ecosystem structure throughout the Brazilian Amazon from Landsat observations and automated spectral unmixing. Earth Interact.
**2005**, 9, 1–31. [Google Scholar] [CrossRef] - Guerschman, J.P.; Oyarzabal, M.; Malthus, T.; McVicar, T.; Byrne, G.; Randall, L.; Stewart, J. Evaluation of the MODIS-Based Vegetation Fractional Cover Product; Commonwealth Scientific and Industrial Research Organization (CSIRO): Canberra, Australia, 2012.
- Guerschman, J.P.; Scarth, P.F.; McVicar, T.R.; Renzullo, L.J.; Malthus, T.J.; Stewart, J.B.; Rickards, J.E.; Trevithick, R. Assessing the effects of site heterogeneity and soil properties when unmixing photosynthetic vegetation, non-photosynthetic vegetation and bare soil fractions from Landsat and MODIS data. Remote Sens. Environ.
**2015**, 161, 12–26. [Google Scholar] [CrossRef] - Qu, L.; Han, W.; Lin, H.; Zhu, Y.; Zhang, L. Estimating vegetation fraction using hyperspectral pixel unmixing method: A case study of a karst area in China. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens.
**2014**, 7, 4559–4565. [Google Scholar] [CrossRef] - Mu, X.; Hu, M.; Song, W.; Ruan, G.; Ge, Y.; Wang, J.; Huang, S.; Yan, G. Evaluation of sampling methods for validation of remotely sensed fractional vegetation cover. Remote Sens.
**2015**, 7, 16164–16182. [Google Scholar] [CrossRef]

**Figure 2.**Photos of major vegetation structure and characteristics in the Otindag Sandy Land: (

**a**) open woodland; (

**b**) grassland encroached by shrub; (

**c**) degraded grassland; and (

**d**) high percentage of non-photosynthetic vegetation (NPV).

**Figure 3.**Endmember spectra of (

**a**) Photosynthetic vegetation (PV); (

**b**) Non-photosynthetic vegetation (NPV) and (

**c**) Bare soil used in Spectral Mixture Analysis (SMA), the filled squares are the average values of different types.

**Figure 4.**f

_{pv}, f

_{npv}and f

_{soil}estimation based on: (

**a**) Spectral Mixture Analysis (SMA); (

**b**) Multi-Endmember Spectral Mixture Analysis (MESMA) and (

**c**) Automated Monte Carlo Unmixing Analysis (AutoMCU).

**Figure 6.**f

_{pv}(

**a**); and f

_{npv}(

**b**) distribution and their average values relative to unmixing iterations using Automated Monte Carlo Unmixing Analysis (AutoMCU).

**Figure 7.**Photosynthetic vegetation (PV), Non-photosynthetic vegetation (NPV) and bare soil endmember variability in Multi-Endmember Spectral Mixture Analysis (MESMA).

**Figure 8.**fpv and fnpv differences based on Spectral Mixture Analysis (SMA), Multi-Endmember Spectral Mixture Analysis (MESMA) and Automated Monte Carlo Unmixing Analysis (AutoMCU).

Sensor | Acquisition Date | Spectral Bands |
---|---|---|

WFV3 | 31 July 2014 | 450–520 nm (Blue) |

WFV3 | 31 July 2014 | 520–590 nm (Green) |

WFV4 | 31 July 2014 | 630–690 nm (Red) |

WFV4 | 31 July 2014 | 770–890 nm |

WFV2 | 4 August 2014 | (Near infrared) |

**Table 2.**Characteristics of sample plots. Values shown for f

_{pv}, f

_{npv}, and f

_{soil}are average ± standard deviations.

Sample Plots Type | Numbers | f_{pv} | f_{npv} | f_{soil} |
---|---|---|---|---|

Open woodland | 12 | 0.44 ± 0.20 | 0.24 ± 0.10 | 0.30 ± 0.22 |

Grassland encroached by shrub | 40 | 0.52 ± 0.15 | 0.14 ± 0.12 | 0.32 ± 0.17 |

Grassland | 52 | 0.57 ± 0.23 | 0.14 ± 0.10 | 0.29 ± 0.24 |

Total | 104 |

Unmixing Approach | PV | NPV | BS | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

R^{2} | RMSE_{T} | RMSE_{OW} | RMSE_{GS} | RMSE_{G} | R^{2} | RMSE_{T} | RMSE_{OW} | RMSE_{GS} | RMSE_{G} | R^{2} | RMSE_{T} | RMSE_{OW} | RMSE_{GS} | RMSE_{G} | |

AutoMCU | 0.49 * | 0.17 | 0.14 | 0.16 | 0.19 | 0.49 * | 0.09 | 0.13 | 0.10 | 0.07 | 0.48 * | 0.20 | 0.13 | 0.22 | 0.20 |

MESMA | 0.48 * | 0.21 | 0.15 | 0.20 | 0.23 | 0.11 | 0.24 | 0.21 | 0.25 | 0.25 | 0.15 | 0.21 | 0.21 | 0.18 | 0.23 |

SMA | 0.47 * | 0.27 | 0.19 | 0.25 | 0.29 | 0.41* | 0.20 | 0.25 | 0.18 | 0.21 | 0.47 * | 0.17 | 0.17 | 0.18 | 0.16 |

_{T}stands for total RMSE, RMSE

_{OW}stands for the RMSE of open woodland samples, RMSE

_{GS}stands for the RMSE of grassland with shrub samples, RMSE

_{G}stands for the RMSE of pure grassland samples. * p < 0.0001.

# | Reference | Source Data | Study Region and Area | Study Period | Approach | Validation Points | RMSE of f_{pv} | RMSE of f_{npv} |
---|---|---|---|---|---|---|---|---|

1 | Guerschman et al. (2012) [42] | MODIS NDVI and the ratio of MODIS bands 7 and 6 | Australia; ~7.7 × 10^{6} km^{2} | 2000–2010 | SMA | 567 | 14.7% | 20.5% |

2 | Okin et al. (2013) [26] | MODIS | Australia; ~150 km^{2} | April, July and October 2010 | SMA, MESMA | 27 | 7%–23% | 12%–29% |

3 | Guerschman et al. (2015) [43] | Landsat and MODIS | Australia; ~7.7 × 10^{6} km^{2} | 2000–2013 | SMA | 1171 | 11.2%–11.9% | 16.2%–17.4% |

4 | Current study | GF-1 WFV | Otindag Sandy Land of North China; ~3.0 × 10^{4} km^{2} | Peak growing season, 2014 | SMA, MESMA and AutoMCU | 104 | 17%–27% | 9%–24% |

© 2016 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Li, X.; Zheng, G.; Wang, J.; Ji, C.; Sun, B.; Gao, Z.
Comparison of Methods for Estimating Fractional Cover of Photosynthetic and Non-Photosynthetic Vegetation in the Otindag Sandy Land Using GF-1 Wide-Field View Data. *Remote Sens.* **2016**, *8*, 800.
https://doi.org/10.3390/rs8100800

**AMA Style**

Li X, Zheng G, Wang J, Ji C, Sun B, Gao Z.
Comparison of Methods for Estimating Fractional Cover of Photosynthetic and Non-Photosynthetic Vegetation in the Otindag Sandy Land Using GF-1 Wide-Field View Data. *Remote Sensing*. 2016; 8(10):800.
https://doi.org/10.3390/rs8100800

**Chicago/Turabian Style**

Li, Xiaosong, Guoxiong Zheng, Jinying Wang, Cuicui Ji, Bin Sun, and Zhihai Gao.
2016. "Comparison of Methods for Estimating Fractional Cover of Photosynthetic and Non-Photosynthetic Vegetation in the Otindag Sandy Land Using GF-1 Wide-Field View Data" *Remote Sensing* 8, no. 10: 800.
https://doi.org/10.3390/rs8100800